Re: [PATCH 0/4] OMAP4: DSS2: Adding fclk support for DPI interface

["Cousson, Benoit" <b-cousson@ti.com>]

Hi Murthy,

On 2/16/2011 9:57 AM, Murthy, Raghuveer wrote:
> On Monday 14 February 2011 09:32 PM, Valkeinen, Tomi wrote:
>> Hi,
>>
>> On Thu, 2011-02-03 at 07:56 -0600, Murthy, Raghuveer wrote:
>>> - Adding dss_feature for DPLL fclk
>>> - Enabling pixel clock generation for DPI interface
>>
>> A bit more description what the patch set is about would be nice. Also,
>> one line patch descriptions are a bit too short for anything else than
>> the most trivial patches.
>>
>> Now to the actual patch contents:
>>
>> DPLL is not a feature of the DSS, and I don't think we should have
>> dss_features for that. In fact, I think the whole DPLL code should be
>> moved from DSS to somewhere under arch/arm.
>>
>> In a perfect world DSS could just set the dss_fck to whatever rate it
>> requires, but as the clock rate can only be set to certain rates, and we
>> need a precise control for the rate, some other method has to be in
>> place.
>>
>> I am not sure what this method should be. Perhaps there is something in
>> the clock framework that could help us here, or perhaps we just need a
>> bunch of function pointers in the DSS's platform data which can be used
>> to configure the clock.
>>
>>    Tomi
>>
>>
>
> Hi Paul, Benoit,
>
> DPLL_PER post divider output for DSS core functional clock can be
> changed in OMAP3xxx and OMAP4430, based on the requested pixel clock for
> a given display resolution.
>
> Additionally, the number of dividers available for DPLL_PER post
> dividors for DSS has increased from 16 to 32, from OMAP3630 onwards.
>
> Both these are added as dss_features.
>
> Given the above comments from Tomi, can the same be included as part of
> the clock framework?

Yes, I do agree with Tomi, this code has nothing to do in the DSS driver.
We do have similar issue with clock rate that can be changed by 2 
successive clock nodes (DPLL and HS divider for example).

But it might be tricky to handle in a generic manner. We need to tie 
these two clock nodes and thus ensuring that there is no other 
descendant for the parent node and then force a parent rate change if 
the clock rate cannot be achieve by the descendant.

Maybe Paul has some idea about that.

Regards,
Benoit